Contest
week-02 · tasks & rules
Open any row for the full spec, examples, and what the grader checks.
All tasks
| # | Task | Grading | Pts |
|---|---|---|---|
| 1 | Task 01 — Hello Container | — | 12 |
| 2 | Task 02 — Context Leak | — | 10 |
| 3 | Task 03 — The Flapping Stack | Live SSH | 14 |
| 4 | Task 04 — Misleading Redis Error | Live SSH | 14 |
| 5 | Task 05 — Migration Before Ready | Live SSH | 15 |
| 6 | Task 06 — Slim the Image | — | 12 |
| 7 | Task 07 — Permission Rabbit Hole | Live SSH | 16 |
| 8 | Task 08 — Data Ghost Volume | Live SSH | 22 |
| 9 | Task 09 — Exec Rescue (Five Bugs) | Live SSH | 14 |
| 10 | Task 10 — Container Doctor | — | 14 |
| 11 | Task 11 — 2 AM Incident | Live SSH | 24 |
| 12 | Task 12 — Full Stack Rescue (Capstone) | Live SSH | 33 |
Containerize, debug, and rescue real Docker stacks. Task specs show symptoms only — no spoilers.
Setup, scoring, and per-task commands live on the Instructions page (not duplicated here).
Open any row in the All tasks table below for the full spec, grader rubric, and How to evaluate block.
Code of honour
Solve them yourself. Docs, blog posts, and AI assistance are fine. Copy-pasting another student's files is not.
Scoring & penalties
Standing score
Your leaderboard score is the sum of your best mark on each task across all graded submissions — not your best single upload:
score = Σ (best points on task T)Example: best on task 07 = 9 and best on task 02 = 7 gives a score of 16, even if you submitted many times. Week 2 totals 200 points across 12 tasks.
Partial marks
Many tasks award credit in chunks (CLI shows PART with got/max).
Your standing uses the best partial or full score per task.
Task 12 (capstone) is worth 33 points with multiple graded checks.
Penalties (tie-break only)
Penalties never subtract points. They break ties when two players have the same score. For each task T, penalty on T = max(0, penalty_runs_on_T − 1):
| Penalty runs on T | Penalty |
|---|---|
| 1st | 0 |
| 2nd | 1 |
| 3rd | 2 |
| … | runs − 1 |
What counts as a penalty run:
csot submit ./solutions -t 07— every graded run on task 07 counts.- Full-folder submit — counts only when the task file was present and you had not already reached full marks on that task before this submission. Missing files and already-perfect tasks are ignored.
- Skipped tasks (partial submit filter) and identical re-uploads (deduped) never count.
Live-sandbox time penalty (Week 2)
Week 2's live incidents (tasks 03, 04, 05, 07, 08, 09, 11, 12) behave like a real on-call page: the longer your sandbox container stays running before you fix it, the more points that task loses. Unlike the tie-break penalty above, this does reduce your score — but never below 80% of the task.
- The clock counts container uptime only. Run
csot challenge stopwhen you step away and the clock pauses — idle time is free. - Only one sandbox runs at a time: starting/switching to another task auto-stops the previous one and pauses its clock.
- Submit-only tasks (01, 02, 06, 10) have no live sandbox and are never time-penalised.
| Container uptime on T | Points lost |
|---|---|
| under 30 min | 0% |
| 30 min | 2.5% |
| + every extra 20 min | +2.5% |
| 170 min and beyond | 20% (capped) |
Check your per-task VM time and current penalty any time with:
csot challenge list -w week-02
(shows a VM time and Time pen. column per task, plus your total uptime for the week).
Tie-break order
Higher score (after time and AI penalties) → fewer total attempt-penalties → earlier last graded submission.
AI usage policy
Using AI tools (ChatGPT, Copilot, Claude, etc.) to help you understand the problem or to write your own fix is perfectly fine — that is how you learn.
Giving an AI agent direct SSH access to your sandbox is penalised. AI agent access is detected automatically. Each detected session costs 2% of that task's points (no floor — repeated abuse compounds).
| Allowed ✓ | Penalised ✗ |
|---|---|
| Asking AI to explain a concept or error message | Running ssh user@host "command" via an AI agent script |
| Generating a fix yourself and applying it over SSH | Piping SSH stdin/stdout into an LLM automation loop |
| Any interactive, human-driven terminal session | Automated shell sessions |
AI agent access is detected automatically. Check your AI flag count with
csot challenge list -w week-02.